The point-spread function (PSF) of an adaptive optics (AO) system is often poorly known. This ignorance can lead to significant systematic errors in photometry. Since the degree of AO correction is sensitive to the observing conditions: seeing, wind speed, brightness of the wavefront reference, etc., it would be desirable to estimate the PSF from the data themselves. We have developed a method to estimate the PSF delivered by an AO system in the case where the scene consists of a crowded star field. We model the modulation transfer function (MTF) of several key components of the imaging system. The power spectrum of the image, even a dense star field, can then be used to constrain our model, which in turn can be used to reconstruct the PSF. We show that the method yields reasonable fit parameters and a useful approximation to the PSF when applied to data from laser guide star (LGS) AO system at the Keck Observatory. Comparison of Keck LGS/AO and HST/NICMOS data shows that photometric accuracy of a few percent can be achieved for data with Strehl ratios as low as 4%.